further responses should go to tech-net.

>> 	this could mean that the outgoing interface on "drifter" is not
>> 	initialized properly and cannot generate packets right after the
>> 	initialization, or whatever.  it basically is a timing issue - there
>> 	are some delay factor implemented in the kernel code, but for some
>> 	of the drivers/ethernet cards they are not enough.
>This sysctl solves my problems, so it seems indeed timing related:
>	sysctl -w net.inet6.ip6.dad_count=3
>Setting it to three makes duplicate detection work between tlp0 and
>ex0, but not for tlp0 and ep0 (latter on slow i486 laptop). Setting it
>to 5, makes duplicate detection work for the latter to.

	that would work, but they are just workaround.  i don't think it
	a serious problem - we will not see too many address conflicts,
	as we advocate stateless autoconfiguration (= almost no possibility
	in interface ID) anyways.

	for post-1.5, i'd like to see some in-kernel indication from drivers
	that they are really really ready, but normally OS have no idea about
	it.

itojun